<p>To improve the accuracy of localization in Internet of Things (IoT) applications, we propose a Hybrid Adaptive Learning Rate Optimization (HALRO) system. HALRO improves model training by a novel hybrid learning rate scheduling strategy that dynamically adjusts the learning steps using the integration of schedules like cosine annealing, power decay and exponential decay. It enables HALRO to respond more effectively to the varying data characteristics for improved convergence and generalization. The system is theoretically supported by convergence and generalization bounds. It is empirically validated using multiple deep learning models including CNN, RNN and LSTM with a variety of optimizers. HALRO also offers higher positioning accuracy particularly in environments that are based on Received Signal Strength Indicator (RSSI), a method that estimates device location. Extensive simulations confirm that HALRO is a scalable and reliable solution for real-time IoT localization. Owing to its adaptability and performance, it is an optimal choice for broader industrial deployment, with applications in logistics, smart cities and other machine learning tasks beyond IoT.</p>